XVERSE Deep Dive: A Simulation-World Route to Physical AI Built on the XVERSE Model Family, an Edge-Cloud 3D Engine and the Physics-Based MotionGen Model
XVERSE pursues physical AI through a cognition-plus-perception dual track: an XVERSE model family spanning 7 billion to 255 billion parameters, an edge-cloud 3D engine, and the physics-based MotionGen motion model, with the VISION WALK VR large-space brand opening 150-plus China stores within a year.
Executive Summary
XVERSE (Shenzhen Yuanxiang Information Technology Co., Ltd.) operates on a cognition-plus-perception dual track: the cognition line is the XVERSE model family (7B to 255 billion parameters, including what was at release the largest open-source MoE model in China), the perception line is an edge-cloud collaborative 3D engine and 3DGS scene reconstruction, and the behavior line is MotionGen, a text-to-motion model released in August 2024. Commercially, the VISION WALK VR large-space brand has opened more than 500 experience stores across over 140 Chinese cities (an older caliber cited 150 stores within a year of its July 2024 launch), and the Saylo roleplay app has accumulated over 20 million registered users (older caliber 3 million). Founded in early 2021, the company has raised over US$200 million from Tencent, Gaorong (GGV Capital affiliate), 5Y Capital, Hillhouse, Sequoia China, Temasek and CPE, with a reported valuation above RMB 10 billion.
This report relies on public reporting and company statements. The company is private and discloses no revenue or margin figures; store counts, user metrics and training-efficiency numbers are company or media claims, and each is graded by source in the Discrepancies section.
XVERSE occupies an unusual position in the physical-AI landscape: it builds no solvers, no surrogate models and no robot hardware. It approaches the field from three directions — reconstructing worlds (3D), driving behavior (motion generation) and interpreting intent (large models). These are upstream capabilities for embodied AI and simulation worlds: 3DGS reconstruction supplies synthetic scenes for embodied training, physics-driven motion generation can supply behavior data for robot learning, and the large model handles orchestration. The assessment: a company with real technical assets, a physical-AI positioning that remains largely narrative, and commercial traction concentrated in entertainment and cultural tourism.
Company Overview
XVERSE was founded in early 2021 in Shenzhen by Forrest Yao (Yao Xing), former Tencent vice president and founding head of Tencent AI Lab and Robotics X, whose teams built the exceptional Go AI Fine Art (Jueyi) and the game AI Juewu; he serves on China's Ministry of Science and Technology strategic advisory committee for AI. Co-founder and CTO Xiao Zhili previously served as general manager of Tencent's infrastructure group and is a specialist in cloud storage and rich-media transmission. The team numbers roughly 200 to 300, with more than 75 percent in product and research roles, drawn from Tencent, Baidu, IBM and McKinsey.
The founding premise came from reinforcement learning: intelligent agents need an environment in which to explore, so building the digital world comes first. That logic explains why a large-model company treats the 3D engine as the other half of its core business.
Three milestones define the trajectory: aggressive open-source releases of the XVERSE model family starting in 2023 established technical credibility; the VISION WALK retail rollout from July 2024 answered the commercialization question of how to reach users before headsets are widespread; and MotionGen in August 2024 added motion generation to the product line.
Technology
Cognition Line: The XVERSE Model Family
The release cadence was dense: a 7-billion-parameter model in August 2023, 13 billion in September, 65 billion (XVERSE-65B) in November, and 30 quantized variants in December. XVERSE-Long-256K followed in January 2024 as the open-source model with the longest context at the time (256K tokens, roughly 600,000 Chinese characters of input). The vision-language model XVERSE-V arrived in March and April 2024. In September 2024 the company released XVERSE-MoE at 255 billion total parameters, then the largest open-source MoE model in China; a general-entertainment MoE base model followed in November. Official specifications for XVERSE-65B: 2.6 trillion pretraining tokens, 40-plus languages, 16K context.
Training efficiency is the emphasized engineering capability: a self-built cluster with a peak GPU utilization of 58.5 percent and an effective training rate of 98.6 percent, using FlashAttention2, 3D parallelism, virtual pipelining, a 4D-topology communication scheme and computation overlap, with a claimed 30 percent reduction in training time. These figures have no third-party benchmark, but the specifics make the claims checkable in principle — a contrast with peers who publish models but not cluster statistics.
Perception Line: Edge-Cloud 3D Engine and 3DGS
The 3D engine differentiates on edge-cloud collaboration: lightweight rendering on the device, heavy rendering in the cloud. Official figures cite co-server capacity on the order of 2 million concurrent users and 1 million online. For offline venues where many users walk the same virtual space at low latency, and for phone-class or standalone-headset devices, this is a practical necessity.
On scene reconstruction, XVERSE adopted 3D Gaussian Splatting roughly three months after the paper's release and shipped a free UE5 plugin, XScene-UE Plugin, covering a one-stop flow from phone capture through 3DGS generation to editing and interactive development inside Unreal Engine 5. For an Apple Vision Pro virtual house-tour project the company specifies: scenes up to 100,000 square meters, 8K resolution (against an industry norm it cites as 4K), and 6DOF movement. The physical-AI relevance is direct: reconstructed real scenes are raw material for embodied-AI synthetic data and simulation-based testing.
Behavior Line: MotionGen and Physics-Driven Motion
MotionGen was released in August 2024, described by the company as China's first 3D AIGC motion-generation model: a short text instruction yields realistic, fluid and complex 3D motion, from basic walking to intricate limb movement. Mainstream motion-synthesis approaches fall into three classes — keyframe animation (hand-made, high quality, high cost), motion-capture retargeting (realistic but limited by capture sessions), and generative synthesis (flexible but prone to foot sliding, ground penetration and floating). Physics-driven means the output must satisfy constraints of center of mass, contact, friction and joint limits; the common industry route treats a physics simulator (MuJoCo, Isaac) as a constraint or post-processor.
MotionGen's implementation details — whether a physics simulator participates, physical-plausibility metrics, generation speed — are undisclosed. Its link to embodied AI is indirect: motion data can feed robot behavior training, but the company discloses no robotics engagement.
Technical Shortfalls
First, the synergy of the three lines is unverified: whether the large model, the 3D engine and MotionGen share representations or reinforce each other is nowhere documented. The company's vision of a large model generating character motion, facial expression and lip sync in real time remains aspirational.
Second, MotionGen ships no performance metrics and no public benchmark; the first-in-China claim is company-only.
Third, the large model's role in physical AI is an interface layer — interpreting intent and orchestrating tools; language models do not perform physical computation, and the dual-track narrative does not explain how the two tracks couple technically.
Product and Commercial Model
Business Lines and Revenue Structure
VISION WALK is the cash-flow anchor: launched July 2024, more than 500 stores across over 140 Chinese cities (an older caliber cited 150 stores within a year), with overseas expansion ongoing. The content library covers titles such as Lunar Odyssey, Twenty Thousand Leagues Under the Sea, Everlasting Tang Dynasty and Journey to the Core, plus museum and cultural-landmark collaborations. Revenue is per-session or per-time; the competitive levers are content, locations and operations.
B-side 3D content production is the second line: disclosed customers include CCTV, Tencent Music, FAW-Volkswagen, Adidas, Postal Savings Bank of China and the Chinese Academy of Sciences, across virtual tours, cultural-tourism guidance and marketing.
The model family and Saylo form the technology and user asset line: open-source models build reputation, while Saylo (an AI roleplay interactive-fiction app) carries consumer monetization. It launched in Hong Kong, Macao and Taiwan in June 2024 (No. 1 on the traditional-Chinese entertainment charts at peak), the United States in October 2024 (No. 3 on Google Play entertainment popularity), and Japan in January 2025. By July 2025 the company reported over 3 million cumulative users, more than 1 million monthly actives, average retention above 65 percent and 110 minutes of daily use among actives — company figures, but retention and session length, if accurate, are well above category norms.
In June 2026 XVERSE launched CastLoop, an AI interactive film-game platform that combines text- or voice-driven interactive storytelling with 3D scenes, extending the "AI-generated 3D content" product form; revenue and user scale are undisclosed, so this report lists it in the product matrix without counting it in the commercialization score.
Revenue Structure
No revenue split is disclosed. From structure and staffing, VISION WALK and B-side projects plausibly carry most revenue, Saylo adds overseas consumer income, and the large-model business itself likely generates no material revenue — the common pattern for China's open-source model startups.
Commercial Risks
Offline VR went through a shakeout after the pandemic; the ceiling is bounded by locations, ticket price and repeat rates. Opening 150 stores in a year indicates operational and content competence, but the business remains offline entertainment retail. Saylo competes with far larger consumer platforms. In substance the company is a combination of offline entertainment, an overseas AI application, and technology reserves.
Competitive Landscape
By segment: VISION WALK competes with offline immersion brands such as Danghong Qitian and theme-park operators; the 3D engine competes with Unity, Unreal and Cocos domestically; the model family competes with large internet companies and model startups; motion generation competes with capture service vendors and generative-animation tools.
The differentiation claim is simultaneous depth in both AI and 3D: the lead investor's profile calls XVERSE the only company in China with deep positions in both. That uniqueness matters only if the three lines couple — if so, XVERSE gains an end-to-end cost advantage in AI-generated 3D content (interpret intent, generate the scene, drive the characters); if not, it is three small companies stacked together. On present public evidence, synergy remains narrative.
Against physical-AI-native companies such as World Labs (spatial intelligence) and AMI Labs (world models), XVERSE's 3DGS-plus-MotionGen stack is closer to a content-production toolchain. The overlap is in reconstructing and driving worlds; the absence of a simulation core limits its standing as engineering-oriented physical AI.
Financing and Capital Structure
Cumulative funding exceeds US$200 million (Aiqicha aggregation) from Tencent, Gaorong Capital (co-lead of the angel round with continued participation), 5Y Capital, Hillhouse, Sequoia China, Temasek and CPE. Media reporting puts the valuation above RMB 10 billion, with four consecutive Hurun Global Unicorn listings. Round-by-round amounts and the latest valuation are not fully disclosed.
Capital consumption runs on two tracks at once — model training (compute) and store expansion (equipment, content, locations) — plus overseas user acquisition for Saylo. If all three lines continue through 2026, the funding cadence must match the commercialization slope.
DISCREPANCIES
1. Three store-count figures. Original reporting and company statements cite 140-plus cities and 500-plus stores; the lead investor's profile says VISION WALK opened 150-plus stores in China within a year; Aiqicha says 150-plus stores at home and abroad. These may correspond to different points in time (150-plus by mid-2025, 500-plus possibly by 2026), but no public data cross-validates the growth curve. This report uses 150-plus (mid-2025, two consistent sources) as the primary figure and marks 500-plus as company-only. Source grade: MEDIA (two) versus OFFICIAL (one) conflict.
2. The exclusivity of the first-in-China claim for MotionGen. China's first 3D AIGC motion-generation model is a company claim without a defined comparison set (does it include capture-retargeting products or academic models?). Source grade: OFFICIAL (company statement).
3. Cumulative funding and valuation. The over-US$200-million figure comes from Aiqicha as an aggregation; the above-RMB-10-billion valuation comes from Economic Information Daily reporting; round-level amounts and the latest valuation have no official disclosure. Source grade: DB and MEDIA.
4. Training-efficiency figures lack third-party validation. Peak utilization of 58.5 percent, effective training rate of 98.6 percent and a 30 percent training-time reduction are company figures with no stated baselines (denominators, measurement windows). Source grade: OFFICIAL.
5. Time-bound model milestones. Longest open-source context at release and largest open-source MoE in China at release are relative claims already surpassed by later models; the 255-billion figure is total MoE parameters, with active parameters undisclosed. Source grade: OFFICIAL (time-bound).
6. Saylo user metrics. The 3 million cumulative users, 65 percent retention and 110-minute daily session are company figures with no third-party analytics cross-check; the definition of average retention (next-day, 7-day or 30-day) is unstated. Source grade: OFFICIAL.
7. The boundary of full-stack self-development. Whether the training stack builds on open-source frameworks and what pretraining data comprises are undisclosed; full-stack refers to model autonomy from pretraining to deployment, not data provenance or hardware supply. Source grade: OFFICIAL (claim) with missing detail.
8. The 8K comparison for the Vision Pro project. The 8K versus an industry-norm 4K comparison does not specify its referent (streaming resolution, display panel, or render output). Source grade: OFFICIAL.
9. Latest store and city-coverage caliber. 2026 public information shows VISION WALK at more than 500 stores across over 140 cities, a significant upward revision from the 150-store one-year figure used in the main text; the 150-store (mid-2025) figure is an older-timepoint caliber and should be superseded by 500-plus/140-plus with the timepoint noted. Source grade: OFFICIAL (company and investment materials) and MEDIA.
10. Saylo user-scale new caliber. A 2026 disclosure puts Saylo cumulative registered users above 20 million (versus the older 3-million cumulative, over-1-million monthly figure); the user magnitude is revised up about sixfold. The main text uses the 3-million caliber and should be updated to 20-million-plus registered users while retaining the older caliber label. Source grade: OFFICIAL (company) and MEDIA.
11. CastLoop interactive film-game and four-year unicorn. In June 2026 XVERSE launched CastLoop, an AI interactive film-game platform, extending its interactive-content forms; the latest Hurun list shows the company among Global Unicorns for a fourth consecutive year (rather than the previously stated three), so the unicorn-year caliber should be updated. Source grade: OFFICIAL (company) and MEDIA.
Risks and Diligence Gaps
First, the mismatch between core business and physical-AI narrative. Revenue anchors are offline VR entertainment and B-side content projects; physical AI (motion generation, scene reconstruction) is currently capability, not revenue. The investor question is whether this is entertainment cash flow funding technology, or an AI narrative lifting an offline business.
Second, the sustainability of model investment. Training and serving a 255-billion-parameter MoE is expensive and open-source weights resist direct monetization; without visible enterprise revenue, the model line functions as brand expenditure.
Third, synergy without evidence. Whether the three lines share a technology stack, whether MotionGen uses the model's semantic understanding, whether 3DGS feeds motion training data — none has public documentation.
Fourth, the offline ceiling. Store economics are bounded by location, ticket price and repeat rates; overseas expansion raises localization costs.
Fifth, Saylo's regulatory and competitive exposure. Overseas AI-companion apps face minor-protection and content-safety regimes; competition from Character.AI-class products is intense, and retention durability is unproven.
Sixth, key-person dependence. The technical narrative binds tightly to the founder's Tencent record; core-team churn would move both fundraising and customer trust.
Seventh, the diligence checklist. Line-of-business revenue and margins, single-store unit economics (capex, payback), MotionGen technical detail and benchmarks, model training costs, and the latest round's terms and valuation all need priority verification.
Composite Assessment
Technology: 6.5. The model family (7B to a 255-billion MoE, 256K context), the edge-cloud engine (specific 100,000-square-meter, 8K, 6DOF targets), fast 3DGS engineering (a plugin three months after the paper) and MotionGen (text-to-motion) are real capabilities spanning cognition, perception and behavior. Deductions: unproven synergy, no MotionGen metrics, unvalidated efficiency figures.
Business: 7. A 150-store first-year rollout, Saylo's overseas metrics (65 percent retention, 110-minute sessions) and CCTV-class references are checkable commercial facts. Deductions: undisclosed revenue mix, the offline ceiling, acquisition-dependent growth.
Financing: 7. Over US$200 million raised, first-tier institutional backing (Tencent, Hillhouse, Sequoia, Temasek) and unicorn status give ample runway. Deductions: three capital-consuming lines at once, opaque latest-round terms.
Product: 6.5. The matrix runs from models to stores to a consumer app, and the VISION WALK library plus Saylo operations demonstrate execution. Deductions: product dispersion, limited model productization.
Market: 6.5. 3D content production, AI companionship and immersive offline entertainment all have real demand, and cultural-tourism digitalization is a policy tailwind. Deductions: stronger specialist rivals in every segment, and only an indirect physical-AI attribute.
Composite: 6.7. XVERSE is this batch's most concretely commercialized and most indirectly physical-AI company: 3DGS reconstruction and MotionGen are genuine upstream capabilities for the embodied-AI value chain, and VISION WALK and Saylo prove the team can convert technology into revenue — but there is no simulation core and no engineering-simulation market presence. The verdict: XVERSE occupies the consumer and content end of the simulation-world value chain while solver- and surrogate-focused peers occupy the engineering end; if physics constraints in MotionGen harden and motion data opens to robotics, its physical-AI standing would strengthen materially.
References
- XVERSE official site (xverse.cn, accessed 2026): XVERSE-65B specifications (2.6T tokens, 40-plus languages, 16K context); cluster peak utilization 58.5 percent and effective training rate 98.6 percent; edge-cloud rendering with 2-million-class co-server users and 1 million online; Vision Pro house tours (100,000 square meters, 8K, 6DOF)
- Gaorong Capital profile (WeChat, 2025): Yao Xing as former Tencent VP and head of Tencent AI Lab and Robotics X, builder of the Jueyi Go AI and Juewu game AI; Xiao Zhili as co-founder and CTO, ex-GM of Tencent infrastructure; MotionGen released August 2024 as China's first 3D AIGC motion model per the company; XScene-UE Plugin and 3DGS adoption three months after the paper; VISION WALK at 150-plus China stores within a year
- Economic Information Daily (via Sina, 2025): valuation above RMB 10 billion; XVERSE family spanning 7 billion to 255 billion parameters; VP Kong Bo on the cognition-plus-perception route and centimeter-level controller-free gesture interaction
- Aiqicha (accessed 2026): cumulative funding over US$200 million from Tencent, Gaorong, Sequoia China and others; team of 200 to 300 with 70 to 75 percent in R&D; model algorithm registered with the Cyberspace Administration in 2024; open-source XVERSE-MoE-A36B; Saylo topping Hong Kong, Macao and Taiwan entertainment charts with 3-million-plus users; VISION WALK at 150-plus stores home and abroad; customers including CCTV, FAW-Volkswagen, Postal Savings Bank and CAS
- Gaorong profile timeline graphic (2025): August 2023 7B, September 13B, November 65B, December 30 quantized variants, January 2024 XVERSE-Long (longest open-source context), March and April XVERSE-V, September XVERSE-MoE 255B (then largest open-source MoE in China), November entertainment MoE base
- Official site news items (accessed 2026): XVERSE-Long-256K as the first 256K-context open-source model with 600,000-character input; free 3DGS local-training plugin with a one-stop capture, generate and UE5-edit flow
- Saylo operating data (Gaorong profile, July 2025): 3-million-plus cumulative users, 1-million-plus monthly actives, retention above 65 percent, 110 daily minutes among actives; chart positions in Hong Kong, Macao and Taiwan (June 2024), the United States (October 2024) and Japan (January 2025)
- Technical background (public): three classes of motion synthesis — keyframe, capture retargeting, generative — with physics-driven generation requiring center-of-mass, contact, friction and joint-limit constraints, commonly enforced via MuJoCo or Isaac as constraint or post-processing
- Technical background (public): 3D Gaussian Splatting produces high-quality 3D scenes from 2D images with strong quality-efficiency balance; reconstructed real scenes feed embodied-AI synthetic data and simulation testing
- Industry background (public): offline VR economics hinge on content, locations and operations after the post-pandemic shakeout; AI-companion apps face minor-protection and content-safety regimes
- Competitive background (public): the 3D engine market is led by Unity, Unreal and Cocos; China's open-source model startups generally struggle to monetize weights directly
- Financing background (public): Gaorong co-led the angel round with continued participation; four consecutive Hurun Global Unicorn listings; round-level amounts and the latest valuation undisclosed
- XVERSE CastLoop (June 2026): launched CastLoop, an AI interactive film-game platform combining interactive storytelling with 3D scenes, extending interactive-content forms.
- 2026 public information (stores and cities): VISION WALK at more than 500 stores across over 140 cities, a revision from the 150-store one-year caliber used earlier.
- 2026 disclosures (Saylo and Hurun): Saylo cumulative registered users above 20 million (older 3 million); company among Hurun Global Unicorns for a fourth consecutive year (older three).
This report is compiled from public sources and does not constitute investment advice. The company is private; revenue, margins and business-line splits are undisclosed. Store counts, user metrics and training-efficiency figures are company or media claims, unaudited by third parties; uncertainties are flagged throughout.
Related papers
Related tools
Related cases
- TopoDiff: Diffusion-Model Topology Optimization and Its Evaluation Metrics
- HydroGym: A Reinforcement Learning Benchmark Platform for Flow Control
- Building Energy Simulation and Surrogates: The EnergyPlus Case
- Wind Turbine Aeroelastic Simulation and Wake Control: A Verifiable Chain from OpenFAST to Reinforcement Learning
Related reports66
- 51World Deep Dive: A Listed Digital Twin Company Where 51Sim Becomes the First Growth Engine While 51Earth Nearly Vanishes, and Two Irreconcilable Financing Naming Conventions Must Both Be Kept
- AMI Labs: A USD 1.03B Seed Bet on World Models, LeCun's Non-Generative Path and Its Blank Scorecard
- Anshi Pacific (PERA) Deep Dive: A Beijing Domestic-CAE Leader Founded in 2003 Whose PERA SIM General Simulation, AIROM Reduced-Order Models, AgentX Multi-Agent and Jingzhi iGPT/Whale Span from Industrial Simulation to Physics AI
- AspenTech Deep Dive: The Process-Simulation Incumbent Acquired by Emerson and Its Hybrid Industrial-AI Route
- Autodesk Deep Dive: Neural CAD Foundation Models and a $200M World Labs Bet on Physical-World AI, with NAVASTO Adding CFD Surrogate Capability
- Bentley Systems Deep Dive: iTwin Digital Twin and Generative AI Empower Full-Lifecycle Infrastructure Engineering
- BeyondMath Deep Dive: A Foundation Physics AI on the Neural-Operator Route, a $19M Consortium Budget, and Three Misread Facts
- Cadence Deep Dive: System-Level Multiphysics EDA and the Millennium M2000 Supercomputer Anchor Physical AI, with AuraStack Agentic AI and Fidelity GPU CFD as Key Levers
- COMSOL Deep Dive: Four Decades of Multiphysics Solvers, Entering Physical AI via DNN Surrogate-Model Training and All-Physics GPU Acceleration
- Dassault Systèmes Deep Dive: 3DEXPERIENCE and SIMULIA Virtual Twin Physics Behavior Anchor Industrial AI, with LEO Virtual Companion and GPU Explicit Dynamics as Physical-AI Levers
- DP Technology: From the Gordon Bell Prize to National-Level Unicorn Status and the Door of a Hong Kong Listing, the AI-for-Science Leader's Technical Delivery and Commercial Blank
- Emmi AI: From Linz Spin-out to a EUR 330M Mistral Acquisition in 17 Months
- Fainite Deep Dive: A Zurich Physics-AI Simulation Startup Whose Caltech, ETH Zurich, Cambridge and Google Team Uses Physics-Aware Deep Learning to Cut Simulation from Tens of Hours to Seconds, With Only Venture Kick Grants and No Priced Round
- Fysics AI: The PhysX Founder's Second Act, a Domestic Differentiable Physics Engine and Its Ecosystem Play
- Frontop (301313) Deep Dive: From Digital Exhibition Halls to a Physical-AI Data Foundation - FTE Engine, FunSim Simulation and the Gap Between Positioning and Revenue
- Godela Deep Dive: A YC X25 Physics-AI Simulation Startup Whose Ex-Apple, Google and Intel Team Uses Physics-Constrained Surrogate Models to Cut Simulation from Days to Seconds, at Seed Stage with Pilot-Only Access
- HollyWave (688682) Deep Dive: RDSim AI Electromagnetic Simulation and Engineering Agents Close the Simulation-Measurement-Product Loop
- IAT (Alt) Deep Dive: China's First Listed Independent Automotive Design Company Pivots to AI-native R&D with the Yuanfang, Yufeng and Taiyi 3.0 Product System and an NVIDIA Omniverse Partnership, Reporting RMB 975M Revenue and a Widened RMB 250M Net Loss in 2025
- Inductiva Deep Dive: A Closed Case in Open-Solver Orchestration, Eight Premises That Must Be Corrected, and Where Co-Founder and CTO Hugo Penedones Went Next
- Intesim (INTESIM) Deep Dive: A Home-grown Chinese CAE Leader Founded in 2009 with Self-developed Solvers for Structure, Fluid, Electromagnetics, Thermal and Acoustics and Multiphysics Coupling, a ~RMB 100M B Round from CNII and Shenzhen Capital, and 2025 Revenue and Bookings Both above RMB 100M
- iSoftStone (301236) Deep Dive: A ChiNext-Listed IT Services Giant Whose TianShu Platform Builds a Three-Layer Physical-AI Stack with Self-Developed TwinHub Digital Twin, FSim Industrial Simulation, CAE Multiphysics and Phyxis Embodied Control
- Jua Deep Dive: The Earth Physics Transformer as a Large Physics Model for Energy Trading
- Keysight (with ESI Group) Deep Dive: Virtual Prototyping and Hybrid AI Bridge Measurement and Simulation, with Natural-Language Assistants in EDA
- Lightwheel Deep Dive: A Physical AI Data and Evaluation Infrastructure Startup Founded by a Former NVIDIA Simulation Lead, Seven Rounds Totalling Roughly CNY 2.65 Billion at a Valuation Above USD 2 Billion, Seats on Both the Newton and EgoVerse International Standards Committees, With Two Incompatible Accounts of Q1 Orders
- Luminary Cloud: The Physics AI Model Factory Backed by NVentures
- MathWorks Deep Dive: MATLAB/Simulink and Simscape Multiphysics Power Physical AI, with R2026a Copilots and an Agentic Toolkit for Edge Intelligence
- Maniformer (Mifeng) Deep Dive: An Agibot-Spun Physical-AI Data Infrastructure Firm Whose Bodyless MEgo Capture Hardware Is in Mass Production and Which Closed Three Rounds in Half a Year Led by China Telecom
- Mistral AI: How Physics AI Moved From Language Models to EUV Lithography and Nuclear Power
- Monolith AI: The British Engineering-AI Firm That Cut Physical Test Counts Instead of Accelerating Simulation, Now Inside CoreWeave
- Monumo Deep Dive: A Cambridge Deeptech Company Redesigning Electric Motors with the Anser AI Engine, Tens of Millions of Multiphysics Simulations per Day, GBP 10.5M Raised in Total, and a Series A Being Raised
- NAVASTO: Moving Solver Insight Onto the Designer's Desk, and How Autodesk Bought That Entry Point
- Navier: Handing CFD Workflows to Agents, a Seed-Stage Bet on AI for Hardware Engineering
- Neural Concept: An EPFL Spin-out Betting on Neural Surrogates, with $138M Raised
- NP Co. (NP Company) Deep Dive: A 2025-founded Paris Inria-born AI-native simulation startup that raised a EUR 6M pre-seed led by Partech with Mistral AI co-founders as angels, claiming a pre-trained physics foundation model at 1,000x speedup — most claims unverified
- NVIDIA Deep Dive: Building the Full-Stack Physical-AI Infrastructure with PhysicsNeMo, Warp, Cosmos World Foundation Models and Omniverse
- Physical Intelligence Deep Dive: A $2.1B Bet on the GPT Moment for Robotics, With pi0.7 Self-Reporting 60-80 Percent Success on Unseen Tasks and Neither Pricing nor Revenue Disclosed for the Pi API
- Physical Superintelligence Deep Dive: A $58M Seed Bet on Virtual Physicists, With Emmy Targeting AI Data Center Multiphysics Optimization as Its First Commercial Proof Point
- PhysicsX: Pretrained Geometry and Physics for Industrial Engineering at a $2.4B Mark
- Prometheus: A $41B Mark, $18.2B Raised, and Not Yet a Product
- PTC Deep Dive: Creo/Onshape and the ThingWorx Digital Thread Power Physical AI, with NVIDIA Isaac Sim Bridging Robotics Design to Simulation
- Qianjue Technology Deep Dive: How a Body-Agnostic Embodied Brain Pairing the Predictive World Model Polibrain with Brain-Inspired Partitioning and Polynomial Representation Reached a Hundred Thousand Cleaning and Service Robots
- Quanscient Deep Dive: Cloud-Native and Quantum-Powered Multiphysics Simulation as the Data Engine for AI-Native Hardware Engineering
- Rescale Deep Dive: An HPC Operating System and Desktop Engineering Workbench in the Cloud, a $115M Series D, and Five Misread Facts
- Schrödinger Deep Dive: Physics-Based FEP+ Simulation and Machine Learning Power Molecular Discovery, with the 2026 Bunsen Agentic Co-Scientist
- Secondmind Deep Dive: A Cambridge Engineering-AI Firm Whose Bayesian Active Learning Cuts CAE Simulation and Test-Bench Runs by up to 90%, with Roughly USD 73M Raised and Mazda Both Customer and Investor
- Shichuang Technology Deep Dive: SupreCAST Cloud-Native Material-Processing Simulation and Industrial Agents Advance the CAE+AI Strategy
- ShonCloud (Shunyun) Deep Dive: Meshless Particle-Method CFD for EV and Nuclear Fluid Multiphysics
- Shuzhi Maijin Deep Dive: A Shenzhen Fully Self-Developed SNN-plus-LBM CFD Startup Claiming Second-Level Results on Hundreds-of-Millions of Grids on Consumer Hardware, With a Tens-of-Millions-RMB Angel Round From a Ministry-of-Finance-Affiliated Fund and an FPGA Migration Plan
- Siemens (incl. Altair) Deep Dive: The 2025 Altair Acquisition Unites Simcenter and Altair, with PhysicsAI Geometric Deep Learning and PhysicsAI Generate as the Physical-AI Core
- Silurian Deep Dive: A Generative Forecasting Transformer Earth-System Foundation Model for Weather Risk on Energy and Infrastructure Assets
- SimScale Deep Dive: A Cloud CAE Platform Built on Core-Hour Subscriptions, EUR 52M of Equity Kept Separate from EUR 25M of Debt, and Five Misread Facts
- Skild AI Deep Dive: A General Robot Foundation Model Incubated at CMU's Robotics Institute, With a USD 1.4B Series C at a USD 14B Valuation, a Conflicting Series B Size, and an ARR Figure That Is a Run-Rate Rather Than Revenue
- Songying Deep Dive: A Huawei-Cloud-Alumni Startup Building a Domestic Physical AI Operating System, With ORCA OS Positioning Against NVIDIA Omniverse, Three Incompatible Financing Narratives, and a Verifiable Third-Party Anchor in a CNY 20 Million Minority Investment
- Sudo Tech Deep Dive: Zero-Real-Data Simulation Training, the Fully In-House Sudo R1 Stack, and an Atomic-Skill Commercialisation Strategy from a Startup Valued Above CNY 20 Billion in 17 Months
- Suochen Technology: The Pain of Becoming a Physics-AI Platform from a CAE Vendor
- Synopsys (Ansys) Deep Dive: The 2025 Ansys Acquisition Turns the EDA Leader into a Silicon-to-Systems Engineering Powerhouse, with Ansys SimAI and the NVIDIA Physical-AI Partnership as the Core Physics-AI Levers
- TenFong Deep Dive: An Academician-Led Chinese Industrial Simulation Software Vendor Whose Official Timeline and Database Records Use Incompatible Round Names, With an Official Financing PDF and an Investment Database Both on File
- Tianfu (Nanjing Tianfu Software) Deep Dive: A 2011-founded Chinese Intelligent-CAE Veteran That Closed a Several-Hundred-Million-RMB Series C, Pivoted to an Industrial World Model in 2024, With Unreconcilable Financing and Valuation Records and No Disclosed Revenue
- Tongyuan Soft Control Deep Dive: MWORKS (Syslab/Sysplorer) as a Domestic MATLAB/Simulink Alternative, with MWORKS AI and the MOGICK Engineering Agent Advancing Home-Grown Physical AI
- Vinci AI: Turning Physics Simulation from a Checkpoint into a Continuous Constraint
- Westlake Robotics Deep Dive: Westlake University's First Robotics Spin-out Builds a Dual-Pretrain Brain-and-Cerebellum Architecture with WR1 Whole-Body Foundation Model and GAE Behaviour Expert
- World Labs: Fei-Fei Li's Spatial-Intelligence Unicorn, Acquired by AMD for $8.2B
- SimWE: From In-House Solvers to a Physical-AI Engine, the Domestic CAE Leader and What Comes Next
- Yunjing Zhifang Deep Dive: A Peking-University-Incubated Cloud-Native GPU CAE Company with the Simversus Platform and the YJ-DORGP Small-Sample Intelligent Solver, Reporting 275 Percent Order Growth, 160 Percent Revenue Growth and Breakeven in 2025
- Zhongke Diwiji (Five Ages) Deep Dive: A CAS-Institute-of-Automation and Tsinghua Spin-out Whose FAM Embodied Foundation Model Uses Action Optical Flow for Cross-Embodiment Generalisation, Reaching 97% Success from 3 to 5 Demos, with EraBrain and Jiling Robots Serving Sinopec, Leapmotor and Midea
- ZWSOFT (688083) Deep Dive: Overdrive Kernel and ZW3D Wukong 2027 Underpin an Integrated CAD/CAE/CAM Stack, with a CAx Agent Platform as the Practical Route to Engineering AI
This report is compiled from public sources. It is for reference only and does not constitute investment advice. Data is current as of the publication date; please refer to the company’s official filings.